Narrowing ??= assignment doesn't work when RHS contains ??

[tomtheisen]

Bug Report

🔎 Search Terms

Control-Flow-Analysis Coalescing

🕗 Version & Regression Information

• The compiler is reporting a type error incorrectly.
• This changed between versions 3.9.7 and 4.0.5.
• Prior to 4.0, the syntax ??= wasn't supported.
• It is unchanged through 4.4.0 nightly.

⏯ Playground Link

Playground link with relevant code

💻 Code

function f1(a?: {}) : {} {
    const n = null ?? {};
    a ??= n;
    return a; // no problem
}

function f2(a?: {}) : {} {
    a ??= (null ?? {});
    return a; // Type '{} | undefined' is not assignable to type '{}'.
}

function identityFunction<T>(a: T): T {
    return a;
}
function f3(a?: {}) : {} {
    a ??= identityFunction(null ?? {});
    return a; // also fine
}

🙁 Actual behavior

On line 9, a type error is reported: "Type '{} | undefined' is not assignable to type '{}'."

🙂 Expected behavior

I would expect that the code should pass type checking. Here's why.

1 . ??= generally supports narrowing types, as shown in the playground link.
2. ?? generally supports nullability narrowing, as shown in the playground link.
3. There's nothing special about a RHS of a ??= that contains a ??.

[danielmayorga]

I believe this is also applicable to the logical OR assignment ||=. The code in the playground mirrors the code and conditions stated above.

Playground link with relevant code

I also believe there are situations where logical AND assignment &&= is inconsistent as well.

Playground link with relevant code

I'm trying to look through the codebase to learn about it and how to fix this issue, and stumbled across these cases as well. Don't want to open a new work item since they are so closely related. Let me know if there is a better procedure to report this.

[fatcerberus]

I believe this is also applicable to the logical OR assignment ||=. The code in the playground mirrors the code and conditions stated above.

If so, this is probably related to #44553.

[AlCalzone]

Duplicate of #41153 and possibly #40494 ?

[tomtheisen]

Seems like this is fixed, as of 5.6.2